SoftmaxLayer.md

CSoftmaxLayer Class

• CSoftmaxLayer Class
  • Settings
    • Normalization area
  • Trainable parameters
  • Inputs
  • Outputs

The class implements a layer that calculates the softmax function on a vector set.

The following formula is applied to each of the vectors:

softmax(x[0], ... , x[n-1])[i] = exp(x[i]) / (exp(x[0]) + ... + exp(x[n-1]))

Settings

Normalization area

// The dimensions over which the vectors should be normalized
enum TNormalizationArea {
    NA_ObjectSize = 0,
    NA_BatchLength,
    NA_ListSize,
    NA_Channel,

    NA_Count
};

void SetNormalizationArea( TNormalizationArea newArea )

Specifies which dimensions of the input blob constitute the vector length:

• NA_ObjectSize - [Default] the input blob will be considered to contain BatchLength * BatchWidth * ListSize vectors, each of Height * Width * Depth * Channels length.
• NA_BatchLength - the input blob will be considered to contain BatchWidth * ListSize * Height * Width * Depth * Channels vectors, each of BatchLength length.
• NA_ListSize - the input blob will be considered to contain BatchLength * BatchWidth * Height * Width * Depth * Channels vectors, each of ListSize length.
• NA_Channel - the input blob will be considered to contain BatchLength * BatchWidth * ListSize * Height * Width * Depth vectors, each of Channels

Trainable parameters

The layer has no trainable parameters.

Inputs

The single input accepts a data blob of any size. The GetNormalizationArea() setting determines which dimensions will be considered to consitute vector length.

Outputs

The single output contains a blob of the same size with the result of softmax function applied to each vector.